A Deep-Learning-Based Observer for State Estimation of Direct Contact Membrane Distillation System Modeled by Differential Algebraic Equations

Due to its high rejection rate and low energy consumption, Direct Contact Membrane Distillation (DCMD) technology is drawing more attention for seawater desalination, to meet the urgent and growing demands for freshwater. State estimation in DCMD system, which is modeled by nonlinear Differential Algebraic Equations (DAE) is crucial for controller design and system's monitoring. In this paper, a novel learning-based observer is proposed for state estimation of the DCMD system. The method consists of an encoder and decoder structure. The encoder allows to transform the DAE system into a linear ODE modulo an output injection in the latent space and the decoder helps in recovering the state estimate from the latent state. First, a brief description of the DCMD system and its DAE model are recalled. Then, the method is presented and illustrated. Explanations on how the learning structures are constructed and trained are provided. Finally, numerical simulations are conducted to illustrate the effectiveness of the proposed learning-based observer design.